NDGF The Distributed Tier1 Site

Overview Reason Design Reality Experience

Reason No Nordic country is big enough to host a Tier-1 All the Nordic countries together are big enough to warrant a Tier-1 6% of ATLAS and 9% of ALICE Tier-1 resources Currently distributed among seven HPC sites from Copenhagen in the south to Umeå in the north Plus Slovenian Tier-2 disk installed technically as NDGF T1 disk

Design Computing on ARC It's a grid, so just run jobs wherever Mostly on shared HPC resources Form the Tier-1 and Tier-2 sites in accounting and BDII Storage in distributed dCache Took a bit of development effort to make it appear as one endpoint Central headnodes, pools out at sites Ops/Devel staff mix Fix software problems in the software projects instead of creating procedures to handle them in operations

Reality - ARC

Reality - Networking The clouds point to where there are dCache pools

Reality - Storage Pools at 7 different sites Namespace, admin, and doors centrally Graphs show a busy Alice day and a calm Atlas day

Experience Distributed storage great for data taking Somewhat less so for data availability Computing on ARC with data staging makes short interventions on storage transparent Distributed funding at least as challenging as distributed storage and computing Professional HPC sites are good for running clusters and storage, professional developers are good at fixing software issues and making users and ops lives happy

Questions?